In this paper, we describe a prototype tele-conference system that sends, receives and presents multimedia data over network for tele-medical diagnosis (MediNet). MediNet supports tele-conference in which doctors can share patient information and 3-D medical images reconstructed from 2-D medical images. Data shared among doctors also includes hand position (pointing), voice, motion images and still images from camera. The system can send and receive various kinds of data in full-duplex at the same time. To present a continuous stream without discontinuation, the receiver uses delay time to avoid trying to play frames that have not arrived. To determine the delay time, the sender sends the recipient dummy frames just after the sender and the recipient are connected. For each dummy frame, the recipient computes the deviation of the actual arrival time from the desired arrival time. The delay time is the maximum deviation time. Then the sender sends the recipient a header and actual data for each frame. The actual data is played at logical time + delay time. If the frame is larger than the maximally allowable size, then the sender segments the frame and the recipient reassembles the segments.